The US Department of Commerce's National Telecommunications and Information Administration (NTIA) is inviting comments on the risks, benefits, and potential policies related to dual-use foundation models. With widely available weights, these open foundation models are fine-tuned by developers using common computing resources. The consultation seeks insights on topics such as the levels of openness in AI models, the wide availability of model weights, the risks and benefits of open versus closed models, and the government's role in setting standards for associated risks.

While NTIA doesn't explicitly define open foundation models, it acknowledges the ecosystem created by companies like Google and Meta, suggesting their open-source models fall under the scope of its definition. The challenge lies in defining 'open' or 'widely available' foundation models, considering varying interpretations among AI companies. NTIA aims to gather more information on the relationship between openness and availability, including details on licenses, distribution methods, and their fit within the understanding of openness.

Why does it matter?

NTIA highlights the potential benefits of open foundational models, emphasising their potential to be shared with those having fewer resources. Small businesses and academic institutions can leverage these models for competitive advantages.

Open models also promote transparency, enabling closer examination by developers in areas impacting rights and safety, such as healthcare and criminal justice. However, concerns include the risk of attacks against proprietary models and the potential misuse of open models for malicious purposes.

The consultation poses key questions on defining wide availability, assessing risks associated with model weights, analysing benefits for competition and innovation, and evaluating the role of government in setting standards. Stakeholders have until 27 March 2024 to provide their comments.